environmental science & policy 14 (2011) 1121–1131
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/envsci
The meaning of system robustness for flood risk
management
Marjolein J.P. Mens a,b,*, Frans Klijn a, Karin M. de Bruijn a, Eelco van Beek a,b
a
b
Deltares, Flood Risk Management, Rotterdamseweg 185, 2629 HD, Delft, The Netherlands
Twente University, Water Engineering and Management, P.O. Box 217, 7500 AE, Enschede, The Netherlands
abstract
article info
This paper explores how the concept of robustness can be defined and made operational for
Published on line 14 September 2011
flood risk management. Decision making about taking actions to reduce the flood risk is
Keywords:
Because decision makers are increasingly aware of uncertainties, new decision-making
System robustness
approaches are being explored. Among other concepts, robustness is put forward by
complicated by uncertainties about the current and future flood probabilities and impacts.
Flood risk
decision makers in the Netherlands as a way to deal with uncertainties. However, they
Resilience
cannot specify what makes a system more robust. This paper defines robustness as the
Vulnerability
ability of a system to remain functioning under disturbances, where the magnitude of the
Regime shift
disturbance is variable and uncertain. We present a conceptual framework for analysing
system robustness, and we suggest how the robustness of flood risk systems can be
quantified. The ideas are tested on the Westerschelde flood risk system in the Netherlands.
This case shows that analysing system robustness provides additional insight into the effect
of flood risk reduction measures on system behaviour.
# 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
Flood risk management has gained increasing attention
throughout Europe over the last decades. Since 1998, floods in
Europe have caused about 700 deaths and more than 25 billion
Euro economic losses (European Commission, 2010). The severe
floods of the Danube and Elbe in 2002, and the floods in Central
and Eastern Europe in 2005 encouraged the implementation of
the European Directive 2007/60/EC. This directive requires
member states to identify the areas at risk of flooding, to
prepare flood risk maps, and to develop flood risk management
plans (European Parliament, 2007). To arrive at these plans,
decision makers need to choose between different measures, for
example between strengthening embankments and building
flood-proof houses, based on decision making criteria.
In common approaches to select or prioritize flood
risk reduction measures, a key decision criterion is the
cost-effectiveness of the measure, which is the ratio between
implementation and maintenance costs and the expected
reduction in flood risk. However, if flood risk is expressed in a
single number, it remains unclear how the flood impact varies
over a range of possible events. Extreme events may lead to an
unacceptably high impact or even a point of no recovery. This
means that a disaster is possible, but it is uncertain when this
will happen.
Recent policy documents about water management in the
Netherlands have introduced the term robustness. They aim
for example for robust nature, robust water systems and
robust decisions (ARK, 2007; National Water Plan, 2008). These
documents show that robustness associates with being
insensitive to uncertainties, but they do not specify what
makes a system more robust nor to what type of uncertainty.
Furthermore, some consider robustness as being equal to
resilience, whereas others associate robustness with very
strong embankments. Implicitly, different interpretations are
* Corresponding author at: P.O. Box 177, 2600 MH, Delft, The Netherlands. Tel.: +31 088 335 8542; fax: +31 088 3358582.
E-mail address: [email protected] (Marjolein J.P. Mens).
1462-9011/$ – see front matter # 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.envsci.2011.08.003
1122
environmental science & policy 14 (2011) 1121–1131
being used. Because a clear definition lacks, there is a danger of
miscommunication among stakeholders.
This paper explores the concept of robustness for flood risk
management. Analysing the robustness of flood risk systems
is expected to provide insight into how the flood impacts vary
over a range of flood probabilities. The approach thus views
flood risk management from a systems perspective. A
comparable approach was adopted by De Bruijn (2005), who
explored the concept of resilience and studied how lowland
river systems react to a range of river discharges. She showed
that a systems approach may widen the range of flood risk
reduction measures considered, because the dynamics of the
system are better understood. The research described in this
paper takes her work as a starting point, and specifically looks
into thresholds in the system response. Thresholds are for
example a drastic change from no response to a large response
(e.g., because a dike breaches), or the point at which a system
is unable to recover from the response to a disturbance. The
latter can be called a regime shift.
This paper first proposes a conceptual framework for
analysing system robustness, in particular of flood risk
systems. The framework is built up from a set of concepts
that describe system behaviour, key characteristics of system
behaviour, and a set of indicators to quantify these characteristics. We exemplify the framework for flood risk
systems.
this paper focuses on system robustness. To the knowledge of
the authors, system robustness has not been made operational for flood risk systems before. Before further defining system
robustness, a definition of flood risk systems is required.
2.2.
Flood risk systems
2.
Framework for analysing system
robustness
Flood risk systems are geographical areas along the coast or
along a river that have the potential to be flooded. The system
comprises biophysical, economic, and social components. The
biophysical subsystem includes the soil, water, air, flora and
fauna, as well as the elevation of the flood-prone area, and the
physical elements to control the flood risk (e.g., embankments
and structures). The socio-economic subsystem includes the
economic value of land uses, the financial situation of the
area, and economic connections to other areas. It also
comprises the people that live in the flood-prone area, among
other things characterized by their age, their health, their
education and their social networks.
Flood risk systems can be disturbed by storm surges at sea,
discharge waves in river catchments, or intensive rainfall
potentially leading to flash floods. These disturbances vary
naturally in time and normally do not cause significant
damage. However, extreme disturbances may cause flooding
of the system and cause damage and/or casualties. Besides by
flooding, the functioning of a flood risk system (working, wellbeing, recreation, production, etc.) may also be disturbed by
other factors, such as an economic crisis, diseases, and wars.
These factors are important to consider, as they may
negatively affect the robustness of the system to flooding.
2.1.
2.3.
Use of the term robustness
The term robust originates from the Latin robustus, meaning
‘strong’ or ‘hardy’. Nowadays, robust means having strength,
being strongly constructed, or performing without failure
under a wide range of conditions, according to the MerriamWebster dictionary. In daily life, robustness is seen as a
desirable characteristic, for instance of electricity networks
that will keep functioning under high demand. In industry and
engineering, a robust design is a product or building that can
survive all kinds of external forces without failing or
collapsing (robust cars, robust buildings, etc.). Robustness is
thus associated with strength and durability.
In the scientific literature, robustness is defined in a variety
of ways for different fields. We roughly distinguish between
system robustness and decision robustness. System robustness is common in the field of engineering and biology, where
it refers to the ability of systems to maintain desired system
characteristics when subjected to disturbances (Carlson and
Doyle, 2002; Stelling et al., 2004; Jen, 2005). Decision robustness
– as a characteristic of policy decisions – is common in policy
analysis and economics, and is used as a criterion for making
decisions under uncertainty (Rosenhead et al., 1972; Bankes,
1993; Lempert et al., 2003; Ben-Haim, 2006). It represents how
sensitive a particular decision is to uncertainties. In other
words, a decision or policy is considered robust when it
performs well under a range of conditions. Although decision
robustness may be very useful in the field of flood risk
management (Hall and Solomatine, 2008; Merz et al., 2010),
Defining system robustness
System robustness is defined as a system’s ability to remain
functioning under disturbances. This implies that information
is needed on how the system responds to different degrees of
disturbance. Therefore, we refer to three streams of literature
that address system behaviour in response to disturbances:
(1) Literature on resilience and resistance of flood risk systems
(De Bruijn, 2004a,b; Klijn et al., 2004), which defines
resistance as the ability to withstand disturbances, and
resilience as the ability to recover from the response to a
disturbance.
(2) Vulnerability literature (Dow, 1992; Turner et al., 2003;
Adger, 2006; Marchand, 2009). Robustness can be thought
of as the flip-side of vulnerability (Gallopin, 2006), referring
to all characteristics of a system that have the potential to
be harmed (Dow, 1992).
(3) Literature on ecological resilience (Holling, 1973; Walker
and Salt, 2006), which defines ecological resilience as the
ability to absorb disturbances without shifting into a
different regime.
All of this literature examines how and why a system
responds to disturbances, but all from different ‘perspectives’.
How these perspectives are linked is visualized in a theoretic
response curve (Fig. 1), which shows the system response as a
function of disturbance magnitude (e.g., the river discharge).
Some find that a disturbance can also be a gradual change, for
environmental science & policy 14 (2011) 1121–1131
1123
Thus to understand system robustness, insight is needed
into the response curve and the recovery threshold. The curve
represents aspects of system robustness, since it visualizes
how the system responds to different degrees of disturbance.
Because of the system’s resistance, the response is zero for a
first range of disturbance magnitudes. Because of the system’s
resilience, it is able to recover from the response to a second
range of disturbance magnitudes. If the curve crosses the
recovery threshold, the point of regime shift is reached.
To arrive at quantifiable indicators of system robustness,
the response curve is described by the following characteristics (adapted from De Bruijn, 2004b):
Fig. 1 – Theoretic response curve, showing system
response as a function of disturbance magnitude,
indicating resistance, resilience, the point of regime shift,
and the recovery threshold.
example sea level rise, to which a system could adapt
(Carpenter et al., 2001; Wardekker et al., 2010). We, however,
consider a disturbance a temporary event. The response refers
to a change in system state, for example the flood impact. For
small disturbance magnitudes, the system may not respond at
all, because it has some degree of resistance. For higher
disturbances, the system may respond and recover, because it
has resilience (De Bruijn, 2004a,b). Resilience is defined as the
ability to respond and recover, which stays closest to the Latin
origin resilire: to jump back. System robustness is a function of
response and recovery.
As mentioned above, vulnerability can be considered the
flip-side of system robustness, since it refers to the impact of a
disturbing event (‘response’) and the ability of individuals or
society to recover from the impact (‘recovery’). Differences
with system robustness may arise when vulnerability is
analysed for only one degree of disturbance. For example in
flood risk management, flood vulnerability maps usually refer
to the potential damage due to a 1 in 100 years river discharge.
If vulnerability is also analysed for the whole range of
disturbances, both the frequent and the rare ones, it can be
considered the flip-side of system robustness.
The third perspective on robustness originates from
ecological resilience literature, which studies the ability of
systems to absorb disturbances without shifting into a
different regime. This implies that the response to a disturbance may be too large to recover from. In ecology, this is
called a regime shift (Scheffer et al., 2001; Scheffer and
Carpenter, 2003), for example the change from a clear-water
lake to an algae-dominated lake. Other examples of regime
shifts can be found in the regime shift database of the
Resilience Alliance (Walker and Meyers, 2004). Studying
regime shifts is considered important for the management
of systems, because if it is understood under which conditions
a regime shift may occur, and what processes drive it, a system
can be managed to persist in the desirable regime (Walker and
Salt, 2006). A regime shift is indicated in Fig. 1 as the point
where the curve intersects the maximum system response
from which recovery is possible (i.e., the recovery threshold,
indicated in the figure with the horizontal dash-dot line). After
a regime shift, the given response curve is not valid anymore.
-
Resistance threshold;
The severity of the response, or amplitude;
The proportionality of the response, or graduality;
The point of regime shift.
The resistance threshold is the point where the response
becomes greater than zero. For flood risk systems, this point
equals the flood protection level. The severity of the response
refers to how far the system has moved away from its normal
situation. The proportionality is introduced to represent the
suddenness of disturbances. It is generally believed that a
more proportional response curve is preferable, in contrast to
one where the response suddenly shows a large increase with
only a small increase in disturbance magnitude (see De Bruijn,
2005). In practice, a more gradual curve implies that floods
occur more frequently and therefore people are expected to be
better prepared. A high resistance threshold usually co-occurs
with a low proportionality, because well-protected areas
attract socio-economic developments. The flood probability
may then be very small, but that one extreme flood will have a
very large impact. A robust system is one where the response
curve stays far from the recovery threshold for a large range of
disturbance magnitudes.
Quantification of the four characteristics together provides
an indication of the system robustness. Fig. 2 visualizes the
steps to be taken when analysing system robustness. Step 1
aims to specify of what system and to what type of disturbance
the robustness is analysed (see also Carpenter et al., 2001).
Section 2.2 already gave a definition of a flood risk system and
the possible types of disturbance. Next, steps 2–4 are explained
for flood risk systems.
Fig. 2 – Conceptual framework for analysing system
robustness.
1124
2.4.
environmental science & policy 14 (2011) 1121–1131
Response curves of flood risk systems
For flood risk systems, the response curve shows to what
extent the socio-economic system is impacted by the disturbance, through inundation of the flood-prone area. Response
variables should be chosen to represent the impact of flooding,
for example the economic damage, the number of affected
persons or the number of casualties. For each response
variable a response curve can be constructed. If there are no
economic or social impacts of flooding, the system response is
considered to be zero. The response depends on the magnitude of the disturbance as well as on the system characteristics. For example, protection measures such as
embankments and storm surge barriers prevent that small
disturbances result in flooding, and obstacles in the floodprone area (such as elevated highways and buildings) may
limit the flood extent for higher disturbance magnitudes. In
natural river valleys, the socio-economic system can be
affected more frequently than in polders with high embankments. In contrast, the response of the river valley will
increase more proportionally with increasing discharges,
since there are no dikes that can breach. It can be expected
that at some point the response (e.g., economic damage) has
reached a maximum, simply because all buildings, infrastructure and crops are totally destroyed. However, the recovery
threshold may be crossed before the maximum system
response is reached, which is the case in Fig. 1.
Although flood risk systems are dynamic systems, the
response curve seems a static system representation. The
purpose of the response curve is to show the effect of flood risk
reduction measures along the full range of disturbances, at a
given moment in time. It is thus not a static curve, but a
snapshot in time. The shape of the response curve may change
over time due to autonomous developments, flood events or
implemented measures.
2.5.
Recovery thresholds of flood risk systems
Recovery of flood risk systems refers to the process of
returning to a normal situation after a flood event. Recovery
involves pumping water out, cleaning the flooded area,
reconstructing houses and other buildings, repairing infrastructure, etc. The long-term impact of a flood event depends
on the time it takes to recover, which in turn depends on the
recovery capacity. We define recovery capacity as the general
socio-economic level of society, referring to system characteristics that influence the ease with which a system recovers.
Recovery capacity is a function of social capital – the ability to
organize repair and reconstruction, and economic capital – the
ability to finance repair and reconstruction (cf. De Bruijn, 2005;
Marchand, 2009). A high recovery capacity implies that only
relatively large impacts will push the system over the recovery
threshold.
As a flood risk system is subject to continuous development, it will not return to the exact pre-disaster state after a
flood event (De Bruijn, 2005). The degree of recovery can be
derived from the system state before and after the event,
indicated by key system characteristics such as population
number, labour force, drinking water availability, and access
to food and electricity. We distinguish between four degrees of
Fig. 3 – System state as a function of time, in response to a
flood event, showing four degrees of recovery.
recovery: recovery, recovery and betterment, partial recovery,
and no recovery (see Fig. 3).
Recovery is when key system characteristics are recovered
to 100% of the pre-disaster situation. In many cases, people
take the opportunity to improve their houses after the
disaster, and policy makers may decide to improve the flood
protection. We call this ‘recovery and betterment’. In extreme
cases, the system does not recover at all, because the incentive
for reconstruction lacks and people do not return; this is called
‘no recovery’, but may also be understood as a regime shift. An
example of such a regime shift is found in the Netherlands,
where a flooding event in the 15th century lead to a change in
land use from agriculture, industry and residential area into
nature. De Biesbosch is a national park in the Netherlands and
one of the few fresh-water tidal areas in Europe (see Fig. 4b). In
the 14th century, De Biesbosch was an urbanized polder, and
part of the dike ring area Groote Waard (indicated with a solid
line in Fig. 4a). The Groote Waard was reclaimed in the end of
the 13th century (Van der Ham, 2003) and consisted of several
agricultural polders and large cities such as Dordrecht and
Geertruidenberg. In 1421, the St. Elizabeth flood flooded almost
the entire polder of the Groote Waard. The polder was still
recovering from this disaster, when a second storm surge
struck in 1424. After the second flood, the community was
unable to rebuild the dikes, and they left the polder as an
estuary. Later in the 17th century, parts of the estuary were
reclaimed again for agriculture. The remaining, unembanked
part of the estuary is now a national park: De Biesbosch
(Fig. 4b). This regime shift may have been caused by a
decreased recovery capacity, for example because the economic resources were already used to rebuild the dikes after
the first flood. Another possible explanation is that societal
support for reconstruction was lacking, because the two floods
happened within only three years. Thus, the system’s ability
to recover reduced after the first flood, which made the system
unable to recover from the second one.
Partial recovery lies in between no recovery and recovery.
For example, New Orleans recovers very slowly from hurricane Katrina and reconstruction is estimated to take 8–11
years (Kates et al., 2006). Currently, society has recovered to
78% of the pre-disaster population and to 87% of the labour
force (GNOCDC, 2010). However, it is still unknown whether it
will reach 100% recovery in the long term, and it can be
environmental science & policy 14 (2011) 1121–1131
1125
Fig. 4 – (a) Map of Groote Waard and surrounding area in 1421 (adapted from Beekman and Schuiling, 1927); (b) current dike
ring areas (numbered), location of national park Biesbosch, and the approximate delineation of the former polder ‘Groote
Waard’.
questioned whether the system will then be comparable to the
pre-disaster situation.
To estimate the potential degree of recovery, we propose to
compare the response curve with a recovery threshold. If the
response exceeds this threshold, the system is not likely to
recover fully. The point of intersection is the point of no
recovery. Each response variable has a different recovery
threshold.
2.6.
Quantifying system robustness
To understand system robustness, we suggested studying the
system response curve, which can be described by four
characteristics. Next, we propose how these characteristics
can be quantified.
2.6.1.
Resistance threshold
The resistance threshold can be quantified by the return
period of the highest disturbance magnitude for which the
response is zero or negligible.
2.6.2.
Severity of the response
The severity of the response refers to the impact of the range
of disturbance magnitudes. It is difficult to estimate the
severity in one number, since it depends on the disturbance
magnitudes, and chosen response variable. For flood risk
systems, De Bruijn (2004a,b) used the expected annual damage
and the expected annual number of casualties to indicate the
response severity. This is equal to the flood risk, and can be
calculated as the area under the response curve, when the
disturbance is expressed in terms of flood probabilities.
2.6.3.
Proportionality of the response
The proportionality of the response refers to the change in
response relative to the change in disturbance magnitude.
According to De Bruijn (2004b), discontinuities in the response
graph point at the possibility of a disaster. She introduced the
measure of graduality. We slightly adapted the original
graduality equation. Whereas De Bruijn calculated graduality
from the resistance threshold until disturbances with a return
period of 1 in 10,000 years, we apply the equation for the whole
response curve, from the 1 in 1 year disturbance until the point
of no recovery. In this way, it includes the suddenness of the
response when the resistance threshold is exceeded, and it
has a clear maximum.
Graduality is calculated by dividing the curve into N
sections, and expressing the disturbance increase and
response increase of each section as fractions of the total
range considered (see Eq. (1)). The fractions are then summed
and divided by 2 to obtain a value between 0 and 1. A value of 1
indicates that the increase of damage is exactly proportionate
to the increase of disturbance. A value close to zero indicates
that the total damage increase occurs at once.
N 1 X
DDn
DRn
G¼1 2 n¼1 ðDmax Dmin Þ ðRmax Rmin Þ
(1)
G = graduality (–)
DDn = change in disturbance for section n (Dn Dn1)
DRn = change in response for section n (Rn Rn1)
Rmax = Response at point of no recovery
Dmax = Disturbance magnitude that corresponds to Rmax
Dmin = Disturbance magnitude with a return period of one
year
Rmin = Response that corresponds to Dmin
N = total number of sections
2.6.4.
Point of no recovery
To arrive at a measure for the point of no recovery, we need
recovery thresholds for each relevant response variable. Little
is known about this point, and more likely it is not a distinct
point but instead a gradual process or an ‘area of no recovery’.
Still, studying the response in relation to a potential threshold
is expected to provide insight into the scale of a disaster. As a
first start, we suggest the following recovery thresholds for
four response variables:
(1) Number of casualties > 10% of the region’s inhabitants;
1126
environmental science & policy 14 (2011) 1121–1131
(2) Number of displaced people > 30% of the region’s inhabitants;
(3) Total economic damage > 50% of the country’s Gross
Domestic Product (GDP) or a state’s Gross State Product
(GSP);
(4) Direct economic damage > 100% of the available public
financial resources.
If any of these thresholds is exceeded, we assume that
recovery will be very difficult and that a regime shift may be
induced. Thus, the point of no recovery can be quantified by
the return period of the disturbance magnitude at which the
response curve intersects with the assumed recovery threshold.
The thresholds are arbitrarily chosen and should be
underpinned by empirical research. Also, it can be disputed
on what scale the thresholds should be analysed. For example,
the number of casualties can be compared to the number of
inhabitants in the country or to those in the flood-prone area.
This will yield different results. Nevertheless, the point of no
recovery is considered to be relevant when comparing the
effects of different flood risk reduction measures.
The first threshold is for the number of casualties, and
proposed at 10% of the number of inhabitants. If a large
percentage of inhabitants deceases, a significant number of
houses may not be rebuild. Likewise, if a large percentage of
the people are displaced, many of them will potentially not
return to the area. The second threshold is thus based on the
number of displaced people relative to the number of
inhabitants. The more inhabitants are displaced for a longer
period, the harder recovery will be. This threshold is chosen at
30%, which comes close to the percentage of displaced people
from Zeeland (the Netherlands) during the flood disaster of
1953. As New Orleans shows, infrastructure and essential
services (electricity, gas, public transportation, schools,
hospitals, and food stores) will not be restored fully as long
as people have not returned (Kates et al., 2006).
The third threshold is based on the total economic damage
relative to the country’s or region’s Gross Domestic Product
(GDP) (see Arakida, 2006). For example, the economic impact of
the 1953 flood disaster in the Netherlands was estimated at 6%
of the national GDP (Klijn and De Grave, 2008). It had no
significant effect on the long-term economic development of
the Netherlands, although the regional societal impact was
considered unacceptable. To compare, the economic impact of
hurricane Katrina is estimated at 81 Billion US Dollar (Knabb
et al., 2006), which is about 44% of the 2005 Gross State Product
of Louisiana ($183 billion) (BEA, 2009). Therefore, we suggest
considering a critical threshold of 50%.
The fourth threshold that we propose is based on the direct
damage relative to the available public financial resources.
Mechler et al. (2006) use the term ‘financing gap’ for the lack of
financial resources to restore assets lost due to a natural
disaster. They assume that public authorities are obliged to
repair housing stock, public infrastructure and relief to the
affected population. To estimate the available financial
resources, data is needed on tax base, budget deficit, internal
and external debt, international loans, aid from other
countries or regions, etc. We suggest considering a critical
threshold of 100%.
3.
Robustness of the Westerschelde flood risk
system and alternative configurations
This case study aims to test the robustness analysis
framework and to explore the robustness of different system
configurations. To be able to calculate the robustness
indicators, we require a hydrodynamic model and a damage
model. Since we modelled the Westerschelde estuary before
(De Bruijn et al., 2008), a range of flooding simulations and the
resulting flood depth maps were available. We used the
Netherlands’ Standard Damage and Casualties model (Kok
et al., 2005; Egorova et al., 2008) to estimate the economic
damage and number of affected persons. The frequency curve
of maximum water levels at Vlissingen was taken from a study
by IMDC (2005), and extrapolated to a water level of 6.6 m
+MSL, which corresponds to the 1 in 10,000 years water level at
Vlissingen in the year 2100 (according to the worst climate
change scenario).
3.1.
Step 1: system definition and type of disturbance
The Westerschelde is a wide estuary in the south-west of the
Netherlands connecting the Schelde river to the North Sea
(Fig. 5). The flood risk system consists of the Westerschelde
estuary and four low-lying polders: Walcheren, Zuid Beveland
West, Zuid Beveland Oost and Zeeuws Vlaanderen. We chose
the western boundary at the city of Vlissingen. Flooding is
assumed to occur from the Westerschelde only, although
Walcheren and Zuid Beveland may also be flooded from the
North Sea and from the Oosterschelde estuary (closed-off from
the North Sea by a storm surge barrier).
The disturbance of interest is a storm surge on the North
Sea. This influences the water level at Vlissingen, which
naturally varies in time due to tidal patterns. A series of
embankments and dunes protects the low-lying polders from
flooding. These are designed to withstand water levels that
occur once in 4000 years on average. At even more extreme
water levels, embankments are expected to fail. Flooding will
cause casualties, damage to buildings and infrastructure, as
well as income losses due to damaged crops. Some damage
occurs indirectly, for example loss of income for the recreation
sector and for companies outside the flooded area due to
flooded road networks.
In addition to the current system configuration, we are
interested in the robustness of three alternative system
configurations. Theoretically, three types of measures can
be identified: those that increase the resistance threshold by
reducing the flood probability; those that decrease the
response or reduce the consequences; and those that increase
the recovery threshold. From the study of De Bruijn et al.
(2008), analysis results of the following measures were
available:
- Strengthening dikes (probability reduction): Flood protection
levels are differentiated between subareas, based on the
expected flood impact. Areas that have a relatively high
economic value are protected up to a water level of 6 m +MSL
at Vlissingen. This implies raising the dikes with 0.4 m.
Other subareas remain protected as in the current configu-
environmental science & policy 14 (2011) 1121–1131
1127
Fig. 5 – Study area with an indication of the flood extent and flood depths.
ration, thus up to a water level of 5.1 m +MSL. Consequently,
fewer dikes will breach at disturbance levels of 5.1–6 m
+MSL.
- Spatial planning (consequence reduction): For example,
building is restricted in areas where inundation depths are
expected to be large, or flood-proof building is enforced. It is
assumed that spatial planning measures will reduce the
economic damage by 30%.
- Storm surge barrier (probability reduction): we assume that a
storm surge barrier is built at Vlissingen. This barrier is
designed to withstand water levels up to 6.2 m +MSL.
3.2.
Step 2: response
Fig. 6a shows the estimated economic damage (in million
Euros) as a function of the water level at Vlissingen, for the
current configuration and the three alternative system
configurations. It shows that the economic damage increases
with an increasing water level at Vlissingen. When more water
is available, this leads to larger water depths and a larger flood
extent. Fig. 6b shows the response curve of the estimated
number of affected persons.
In the ‘strengthening dikes’ configuration, no flooding
occurs until a disturbance of 5.1 m +MSL. When this resistance
is exceeded, only the subareas with a relatively lower
economic value are flooded. At a disturbance level of 6 m,
the better protected areas will be flooded as well. Compared to
the current configuration, ‘strengthening dikes’ has the same
resistance threshold, but a smaller response for disturbances
between 5.1 and 6 m.
In the ‘spatial planning’ configuration, the economic
damage is lower for all disturbances greater than 5.1 m
+MSL. Thus, the resistance threshold is comparable to that of
the current situation. Because spatial planning in this case
only implied flood-proof building and not that people are
motivated to move to less risky places, it only affects the
economic damage and not the number of affected persons.
In the ‘storm surge barrier’ configuration, the resistance
threshold is increased to 1 in more than 100,000 years (i.e., a
water level of 6.2 m). For higher water levels, flooding has the
same pattern as in the reference configuration.
3.3.
Step 3: recovery thresholds
We calculated two of the four proposed recovery thresholds,
and compared each with the corresponding response
curve. The recovery threshold of the economic damage was
proposed at 50% of the GDP. The GDP of the province of
Zeeland was 8418 million Euros in the year 2000 (CBS, 2010).
The threshold of 50% (4209 million Euros) is reached at a water
level of about 6.3 m in all configurations except spatial
planning. Spatial planning does not reach the threshold at
water levels below 6.6 m.
The recovery threshold of the number of affected persons
was proposed at 30% of the number of inhabitants in the
province of Zeeland (year 2000), equalling 112,000 affected
persons. This threshold is indicated in Fig. 6b by the horizontal
dashed line. The reference curve intersects the threshold at a
water level of 6.0 m. ‘Strengthening dikes’ and ‘spatial
planning’ do not influence the point of no recovery. ‘Storm
surge barrier’ moves this point to a water level of 6.2 m +MSL.
From the two quantified thresholds, the recovery threshold
of affected persons can be considered indicative for the point
of no recovery. This point of no recovery is used as the
maximum water level for calculating the graduality, for both
the economic damage curve and the affected persons curve.
1128
environmental science & policy 14 (2011) 1121–1131
Fig. 6 – Response curves for the current and alternative system configurations, including the recovery threshold: (a)
economic damage, (b) number of affected persons.
3.4.
Step 4: quantifying system robustness
The system robustness indicators were calculated following
the methods in Section 2, for both the economic damage and
the affected persons. The severity of the economic damage is
indicated by the expected annual damage (EAD), and the
severity of the number of affected persons is indicated by the
expected number of affected persons (ENAP). Results are listed
in Tables 1 and 2.
As becomes clear from the two tables, all alternative
system configurations show a smaller EAD than the reference,
where ‘storm surge barrier’ shows the largest decrease. Spatial
planning reduces the EAD, but not the ENAP, because floodproof buildings do not affect the number of inhabitants. If a
decision would be based only on a reduction in the EAD and
the ENAP, the storm surge barrier would be the preferred
option. However, the storm surge barrier will probably be the
most costly alternative.
Table 1 – Robustness of the four system configurations, based on the economic damage curve: cur = current situation,
sd = strengthening dikes, spp = spatial planning, ssb = storm surge barrier.
Characteristic
Resistance threshold
Response severity
Response proportionality
Point of no recovery
Point of no recovery
Indicator
cur
sd
spp
ssb
Lowest return period with zero response (1/year)
EAD (million Euro/year)
Graduality (–)
Return period where response equals recovery
threshold ( p/year)
Water level where response equals recovery
threshold (m)
1/4000
0.60
0.32
<1/100,000
1/4000
0.15
0.07
<1/100,000
1/4000
0.42
0.32
<1/100,000
<1/100,000
0.01
0.00
<1/100,000
6.3
6.3
>6.6
6.3
1129
environmental science & policy 14 (2011) 1121–1131
Table 2 – Robustness of the four system configurations, based on the affected persons curve: cur = current situation,
sd = strengthening dikes, spp = spatial planning, ssb = storm surge barrier.
Characteristic
Resistance threshold
Response severity
Response proportionality
Point of no recovery
Point of no recovery
Indicator
cur
sd
spp
ssb
Lowest return period with zero response ( p/year)
ENAP (persons/year)
Graduality (–)
Return period where response equals recovery
threshold ( p/year)
Water level where response equals recovery
threshold (m)
1/4000
20
0.27
<1/100,000
1/4000
4.2
0.07
<1/100,000
1/4000
20
0.27
<1/100,000
<1/100,000
0.3
0.00
<1/100,000
6.0
6.0
6.0
6.2
The resistance threshold is only increased by the storm
surge barrier. The graduality shows the same pattern for the
affected persons curve and the economic damage curve. The
graduality is reduced when dikes are strengthened and when a
storm surge barrier is built. This is because the degree of
sudden increase in response is increased. The graduality of
‘spatial planning’ is equal to that of the reference configuration, although the damage curve shows the smallest absolute
‘damage jump’. This is because the graduality is calculated
relative to the maximum damage, which is also 30% lower
than in the reference. Spatial planning did not influence the
affected persons curve, as explained above.
The point of no recovery is only affected by the storm surge
barrier, which moves the point from 6.0 m to 6.2 m, in the
affected persons curve. The spatial planning configuration
moves the point of no recovery in the economic damage curve
to above 6.6 m, but we considered the affected persons
threshold indicative for the point of no recovery.
3.5.
Case study discussion
With the Westerschelde application, we demonstrated that it
is possible to quantify the response of a flood risk system (in
terms of economic damage and affected persons) for different
disturbance magnitudes. The response curves of three
alternative system configurations were compared with those
of the current configuration, and all robustness characteristics
were quantified. The response curves are considered a good
starting point to discuss system robustness.
From the conceptual framework, we expected that the
point of no recovery would be moved further away by
implementing consequence-reduction measures, since they
would lower the response curve. In the Westerschelde case,
the spatial planning indeed moves away this point in the
economic damage curve. However, in the affected persons
curve, the point of no recovery is only (positively) influenced
by the storm surge barrier, a probability-reduction measure.
This can be explained by the relatively high number of affected
persons that is estimated when the storm surge barrier fails.
Although such an event has a very low probability, the
proposed recovery threshold will be exceeded immediately.
We could say that the storm surge barrier configuration has a
very high resistance and no resilience in terms of affected
persons. In fact, its resistance threshold coincides with the
point of no recovery. However, the economic damage curve is
further away from the recovery threshold. This means that if
the storm surge barrier would fail, the economic threshold will
not necessarily be exceeded; the system has some (economic)
resilience.
One of the proposed indicators is the point of no recovery.
We assumed that if one of the recovery thresholds is exceeded,
this is indicative for the point of no recovery. For example, the
number of affected persons exceeds 30% of the inhabitants at
a disturbance level of 6.0 m (storm surge level). At that point,
the economic recovery threshold is not exceeded yet. As
mentioned before, the point of no recovery is more likely an
‘area’ of no recovery, since recovery is a process and whether
an area will finally recover depends on many factors. The
proposed recovery thresholds are not a hard distinction
between recovery and no recovery. Instead, it shows the scale
of a disaster, and it may indicate how likely it is to recover
fully. However, it should be further discussed how to use the
recovery thresholds in relation to the point of no recovery.
4.
Discussion
In the introduction, system robustness was distinguished
from decision robustness. Because these concepts have been
developed in different fields (biology and ecology versus policy
analysis and decision theory) their meaning and use is
different. Although both concepts refer to the degree of
insensitivity to fluctuations or changes in (external) conditions, three key differences can be identified. First, decision
robustness typically focuses on the consequences of decisions
for the long term, say 50–100 years ahead, whereas system
robustness is analysed for one particular state: a snapshot in
time. Secondly, decision robustness relates to all unknown
future developments for which decision makers want the
decision to be robust (e.g., climate change, economic development and changing societal values). It therefore considers
more than one type of uncertainty. In contrast, system
robustness narrows down to one type of uncertainty, i.e.,
the variability of the most relevant disturbance. Finally, since
decision robustness is about making decisions, it necessarily
takes into account the implementation costs of the decision
options in addition to the benefits. An analysis of system
robustness does not require this type of information; it can be
confined to the most relevant system response and the most
relevant disturbance, while implementation costs play no role.
We believe that both concepts are useful for flood risk
management. We consider them to be complementary, since
they are relevant at a different moment in the decision-making
process. System robustness may help the decision maker to
identify strategies and measures, while decision robustness
may aid to prioritize between identified measures and in
planning their implementation in time. Thus, even for systems
that are not robust, taking no measures may be the most robust
1130
environmental science & policy 14 (2011) 1121–1131
decision from a cost–benefit perspective. To better understand
the benefits of each concept, it is suggested to analyse both
decision robustness and system robustness for one case.
5.
Conclusion
In this paper, we explored several interpretations related to
system robustness, which resulted in a conceptual framework
for analysing system robustness. We defined robustness as the
ability to remain functioning under a wide range of disturbances, and distinguished two main elements: response
and recovery. To analyse robustness, we suggested exploring
the system response and recovery over a range of disturbance
magnitudes. To describe the response curve, we proposed four
characteristics: the resistance threshold, the response severity, the response proportionality and the point of no recovery.
These robustness characteristics may potentially be used as
decision making criteria, but it is still unclear how they can be
combined into a decision making framework.
This paper showed that analysing system robustness
provides insight into how different types of risk reduction
measures affect the behaviour of the flood risk system
(response and recovery). The novelty of this paper is that it
explicitly compares the response curve with a recovery
threshold. It turns out that it highly depends on the recovery
threshold whether probability reduction or damage-reduction
measures will influence the point of no recovery. So, when
aiming for robustness, these thresholds should be investigated. To explore the added value of system robustness for
decision making, and the applicability of the robustness
indicators, more examples are needed.
It is recommended for a next application to compare the
robustness of different system configurations that have an
equal flood risk. In this way, the added value of the additional
indicators can be better evaluated. Furthermore, the recovery
thresholds should be validated by historic examples.
Acknowledgements
The research described in this paper is supported by the
Knowledge for Climate programme (the Netherlands), and the
Deltares research programme 2010. The Westerschelde case
was studied in the context of FLOODsite (European Union, 6th
framework programme, GOCE-CT-2004-505420).
references
Adger, W.N., 2006. Vulnerability. Global Environmental Change
16 (3), 268–281.
Arakida, M., 2006. Measuring vulnerability: the ADRC
perspective for the theoretical basis and principles of
indicator development. In: Birkmann, J. (Ed.), Measuring
Vulnerability to Natural Hazards: Towards Disaster Resilient
Societies. United Nations University Press, Tokyo, Japan.
ARK, 2007. Maak ruimte voor klimaat! Nationale
adaptatiestrategie, beleidsnotitie (English: Make space for
climate, national adaptation strategy). Minsteries van VROM,
V&W, LNW en EZ, IPO, VNG en Uni van Waterschappen (in
Dutch).
Bankes, S., 1993. Exploratory modeling for policy analysis.
Operations Research 41 (May–June (3)).
BEA, 2009. Bureau of Economic Analysis, http://www.bea.gov/
regional/gsp/ (accessed at 11.11.10).
Beekman, A.A., Schuiling, R., 1927. School atlas der Geheele
Aarde (English: School Atlas of the Entire Earth). Uitgeverij
Thieme, Zutphen.
Ben-Haim, Y., 2006. Info-Gap Decision Theory: Decisions Under
Severe Uncertainty, 2nd ed. Academic Press, London.
Carpenter, S., Walker, B., Anderies, J.M., Abel, N., 2001. From
metaphor to measurement: resilience of what to what.
Ecosystems 4, 765–781.
Carlson, J.M., Doyle, J., 2002. Complexity and robustness.
Proceedings of the National Academy of Sciences 99, 2538–
2545.
CBS, 2010. Netherlands Statistics Bureau, http://statline.cbs.nl/
(accessed at 28.09.10).
De Bruijn, K.M., 2004a. Resilience and flood risk management.
Water Policy 6, 53–66.
De Bruijn, K.M., 2004b. Resilience indicators for flood risk
management systems of lowland rivers. International
Journal of River Basin Management 2 (3), 199–210.
De Bruijn, K.M., 2005. Resilience and flood risk management. A
systems approach applied to lowland rivers. Ph.D. thesis.
Technical University Delft, Delft, the Netherlands.
De Bruijn, K.M., Klijn, F., McGahey, C., Mens, M.J. P., Wolfert, H.,
2008. Long-term strategies for flood risk management.
Scenario definition and strategic alternative design.
FLOODsite report T14-08-01 (online: www.floodsite.net).
Dow, K., 1992. Exploring differences in our common future(s):
the meaning of vulnerability to global environmental
change. Geoforum 23 (3), 417–436.
Egorova, R., Van Noortwijk, J.M., Holterman, S., 2008.
Uncertainty in flood damage estimation. International
Journal of River Basin Management 6 (2), 139–148.
European Commission, 2010. http://ec.europa.eu/environment/
water/flood_risk (accessed 10.09.10).
European Parliament, 2007. Directive 2007/60/EC of the
European Parliament and of the Council of 23 October 2007
on the assessment and management of flood risks. http://
eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=CELEX:
32007L0060:EN:NOT (accessed 10.09.10).
Gallopin, C.G., 2006. Linkages between vulnerability, resilience,
and adaptive capacity. Global Environmental Change 16,
293–303.
GNOCDC, 2010. News Release: Facts for Features Hurricane
Katrina Recovery. http://www.gnocdc.org/Factsforfeatures/
HurricaneKatrinaRecovery/index.html (accessed 14.09.10).
Hall, J., Solomatine, D., 2008. A framework for uncertainty
analysis in flood risk management decisions. International
Journal of River Basin Management 6, 85–98.
Holling, C.S., 1973. Resilience and stability of ecological systems.
Annual Review of Ecology and Systematics 4, 1–23.
IMDC, 2005. Faserapport 1: Composietrandvoorwaarden
Maatschappelijke Kosten Baten Analyse Sigmaplan (English:
Combined boundary conditions for the cost benefit analysis
of the Sigma plan), Deelopdracht 1. Ministerie van de
Vlaamse Gemeenschap, Antwerpen, Belgium (in Flemish).
Jen, E., 2005. Stable or robust? What’s the difference?. In: Jen, E.
(Ed.), Robust Design: A Repertoire of Biological, Ecological en
Engineering Case Studies. Oxford University Press, New York.
Kates, R.W., Colten, C.E., Laska, S., Leatherman, S.P., 2006.
Reconstruction of New Orleans after hurricane Katrina: a
research perspective. PNAS 103. (40).
Klijn, F., Van Buuren, M., Van Rooij, S.A.M., 2004. Flood-risk
management strategies for an uncertain future: living with
Rhine river floods in the Netherlands? Ambio 33 (3), 141–147.
environmental science & policy 14 (2011) 1121–1131
Klijn, F., De Grave, P., 2008. Grenzen aan de gevolgen van een
overstroming? Discussiestuk voor Waterveiligheid 21e eeuw
(English: Limits to the consequences of flooding? Discussion
paper for the project Watersafety in the 21st century).
Deltares Report Q4348.50, Delft, the Netherlands
(in Dutch).
Knabb, R.D., Rhome, J.R., Brown, D.P., 2006. Tropical Cyclone
Report Hurricane Katrina 23–30 August 2005. National
Hurricane Center.
Kok, M., Huizinga, H.J., Vrouwenvelder, A.C.W.M., van den
Braak, W.E.W., 2005. Standaardmethode 2005: Schade en
Slachtoffers als gevolg van overstromingen. HKV Lijn in
Water, Lelystad, the Netherlands (in Dutch).
Lempert, R., Popper, S.W., Bankes, S.C., 2003. Shaping the Next
One Hundred Years: New Methods for Quantitative, LongTerm Policy Analysis. RAND, Santa Monica.
Marchand, M., 2009. Modelling coastal vulnerability. Design and
evaluation of a vulnerability model for tropical storms and
floods. Ph.D. thesis. Delft University of Technology, Delft, the
Netherlands.
Mechler, R., Hochreiner, S., Linnerooth-Bayer, J., Pflug, G., 2006.
Public sector financial vulnerability to disasters: the IIASA
CATSIM model. In: Birkmann, J. (Ed.), Measuring
Vulnerability to Natural Hazards: Towards Disaster Resilient
Societies. United Nations University Press, Tokyo, Japan.
Merz, B., Hall, J., Disse, M., Schumann, A., 2010. Fluvial flood risk
management in a changing world. Natural Hazards and
Earth System Sciences 10, 509–527.
National Water Plan, 2008. National Water Plan. Ministry of
Transport, Water Management and Public Works, the
Netherlands.
Rosenhead, M.J., Elton, M., Gupta, S.K., 1972. Robustness and
optimality as criteria for strategic decisions. Operational
Research Quarterly 23 (4), 413–430.
Scheffer, M., Carpenter, S., Foley, J.A., Folke, C., Walker, B., 2001.
Catastrophic shifts in ecosystems. Nature 413, 591–596.
Scheffer, M., Carpenter, S., 2003. Catastrophic regime shifts in
ecosystems: linking theory to observation. Trends in Ecology
and Evolution 18 (12).
Stelling, J., Sauer, U., Szallasi, Z., Doyle, F.J., Doyle, J., 2004.
Robustness of cellular functions. Cell 118, 675–685.
Turner, B.L., Kasperson, R.E., Matson, P.A., McCarthy, J.J., Corell,
R.W., Christensen, L., Eckley, N., Kasperson, J.X., Luers, A.,
Martello, M.L., Polsky, C., Pulsipher, A., Schiller, A., 2003. A
framework for vulnerability analysis in sustainability
science. Proceedings of the national academy of science 100,
8074–8079.
Van der Ham, W., 2003. De Grote Waard, geschiedenis van een
Hollands landschap (English: Landscape History of the Grote
Waard). Uitgeverij 010, Rotterdam. ISBN: 90 6450 506 3.
Walker, B., Meyers, J.A., 2004. Thresholds in ecological and
social–ecological systems: a developing database. Ecology
1131
and Society 9 (2), 3., In: http://www.ecologyandsociety.org/
vol9/iss2/art3.
Walker, B., Salt, D., 2006. Resilience Thinking Sustaining
Ecosystems and People in a Changing World. Island Press,
Washington, DC.
Wardekker, J.A., De Jong, A., Knoop, J.M., Van der Sluijs, J.P.,
2010. Operationalising a resilience approach to adapting an
urban delta to uncertain climate changes. Technological
Forecasting and Social Change 77 (6), 987–998.
Marjolein Mens is a researcher in the field of flood risk analysis
and management, and has been working at Deltares (The
Netherlands) since 2006. She holds a Master’s degree in Hydrology
and Water Management from Wageningen University (The
Netherlands). Currently, she is working on her PhD research
focusing on the analysis of robustness of flood and drought risk
systems, at both Deltares and Twente University. Her work
includes flood hazard assessment, inundation modelling, flood
damage assessment and the development of decision-support
models.
Dr. Frans Klijn is senior specialist in Delta Technology, working at
Deltares (The Netherlands) since 1996. Trained as a physical
geographer and landscape ecologist, he has been working on River
Basin and Coastal Zone Management since 1980. In the nineties,
he played an important role in the societal debate on the management of the Netherlands’ large rivers, which culminated in the
Room-for-Rivers policy. Presently, he co-ordinates research on
climate-proof flood risk management of the Netherlands’ Knowledge for Climate Program.
Dr. Karin de Bruijn is senior researcher in Flood Risk Management,
working at Deltares (The Netherlands) since 1998. Trained as a
hydrologist, she has been working on flood risk analysis and
assessment, flood inundation modelling, flood casualty assessment and the development and assessment of long-term flood risk
management strategies in an uncertain future. In 2005 she
defended her PhD thesis titled ’Resilience and flood risk management, a system’s approach applied on lowland rivers’.
Prof. Eelco van Beek is professor Integrated Water Resources
Management at the University of Twente (The Netherlands), as
well as Water Resources specialist at Deltares (The Netherlands).
Over the last 40 years, he has been actively involved in many water
resources research projects in the Netherlands and abroad with a
focus on drought risk, environmental flow and water allocation.
He was leader of WL j Delft Hydraulics’ team for the important
Policy Analysis for the Water Management for the Netherlands
(PAWN). Presently, he co-ordinates research on climate-proof
drought risk management of the Netherlands’ Knowledge for
Climate Program.